jzOcb

Self Improving Skill — Development skill for Claude Code

Development community

Auto-learn writing rules from human edits.

How to install Self Improving Skill

This entry records only its repository, not the path inside it, so there is no exact command to give. Open jzOcb/self-improving-skill and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Self Improving Skill does

Auto-learn writing rules from human edits. record original → record final → diff → extract rules → update SKILL.md. Works with Claude Code + OpenClaw.

Alternatives in Development

  • Claude Md Management — Tools for managing and improving CLAUDE.md files 14k ★
  • AgentHandover — What if OpenClaw, Claude Code, Codex etc. knew how to do your work without you saying it 600 ★
  • Engage.Report — Execute Phase 8 - Final Reporting and Engagement Closeout 348 ★

README

Self-Improving Skill

让你的 writing style skill 自动从人类修改中学习。

**兼容 Claude Code + OpenClaw (ClawHub)**。`observe.py` 零依赖纯 Python。

原理

AI 写初稿 → 人类改到满意 → diff 两版 → 提取规则 → 更新 SKILL.md → 下次更准

只要两个数据点:**original**(AI 第一版)和 **final**(人类最终版)。中间改了多少轮不管。

安装

# Claude Code
cp -r self-improving-skill ~/.claude/skills/

# OpenClaw / ClawHub
npx clawhub@latest install self-improving-skill
# 或手动: cp -r self-improving-skill ~/clawd/skills/

快速开始

# 1. AI 写完内容后,记录原稿
python3 scripts/observe.py record-original draft.md --skill my-writing-style

# 2. 人类改完确认后,记录最终版
python3 scripts/observe.py record-final final.md --match  --skill my-writing-style

# 3. 提取规则(需要 LLM CLI: claude / llm / 自定义)
python3 scripts/improve.py auto --skill ~/path/to/my-writing-style/

命令

observe.py(零依赖)

命令 功能
record-original 记录 AI 原稿
record-final --match 记录最终版
pending 查看待配对
stats 统计

improve.py(需要 LLM CLI)

命令 功能
extract [--days 7] 提取改进建议
auto 自动提取 + 应用 P0(cron 用)
show 查看提案
apply 应用提案
rollback 回滚

LLM 支持

`improve.py` 自动检测可用 CLI:

  • claude (Claude Code) — 优先
  • llm (pip install llm) — 通用
  • IMPROVE_LLM_CMD 环境变量 — 自定义

`observe.py` 不依赖任何 LLM。

安全

  • 每次 apply 前自动备份 SKILL.md
  • auto 模式只应用高置信度 (P0) 规则
  • 一键 rollback

适用场景

不限于写作。任何 "AI 生成 → 人类修改" 的循环:

  • Writing style(推文、文章)
  • Code review rules
  • Email drafting
  • Translation

License

MIT